Experimental Physics and
Industrial Control System

"Jim Chen" <[email protected]> · Fri, 7 Feb 2014 16:38:52 -0000

Mark and Eric,

I might have pointed to the wrong place. I have deleted everything for the
message queue task as simple as below and it runs fine the first time but
segfault immediately when it waits again:

while(1) {
        /* Wait for event from interrupt routine */
printf("====before\n");
       epicsMessageQueueReceive(intMsgQId, data, sizeof(data));
printf("====queue responsed\n");
}

The debug printout:

====before
....
==== queue responsed
====before
Segmentation fault

This looks like something wrong with the message queue? Anyway I will do
more tests...

Jim  

> -----Original Message-----
> From: Eric Norum [mailto:[email protected]]
> Sent: 07 February 2014 16:27
> To: Jim Chen
> Cc: Mark Rivers
> Subject: Re: asyn driver problem running Linux 3.8.13 with arm processor
> 
> On Feb 7, 2014, at 7:32 AM, Jim Chen <[email protected]>
> wrote:
> 
> > Hi Mark,
> >
> > Last time we discovered that those functions cannot be used in
> > interrupt service routine and I think that makes sense.
> 
> Right ? it?s never possible to block in a call from an interrupt service
routine.
> 
> >
> > The queuetask is a message queue created by epicsMessageQueueCreate
> > waiting interrupt in a thread created by epicsThreadCreate. The
> > strange thing is that in other arch and Linux, it works fine. Is it
> > anything to do with arm processor or the 3.8.13 Linux distro or the gcc
> compiler?
> 
> The calls ?work fine? on other systems I suspect because you?ve been lucky
> ? the request has never had to block.
> It?s likely that something in the architecture/compiler/OS combination of
the
> new system has some significant difference in the execution environment of
> interrupt service routines.   The first that comes to mind is that the
code is
> running on a limited-size interrupt stack and the send call is just taking
up too
> much stack.  Or perhaps something in the new system is resulting in a
> blocking attempt (as some sort of scheduling optimization or something).
> 
> --
> Eric Norum
> [email protected]
> 
> 
> 

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System